← Home

Articles

In-depth analysis of significant developments in agentic commerce and AI infrastructure.

UCP Embedded Protocol Transport: In-Context Commerce for AI Agents

PR #244 introduces ECP — a new transport binding enabling UCP operations within existing channel contexts (chatbots, voice assistants) without browser redirects. Includes reauthorization mechanism for session continuity.

A2A Agent Identity Layer Proposals: WTRMRK and Cross-Org Accountability

Two significant proposals address how A2A agents establish trust across organizational boundaries: WTRMRK watermarking for identity verification and an accountability layer for first-contact transactions.

UCP Identity Linking Revert: Mechanism Registry Pulled Back

Two weeks after merging, Ilya Grigorik reverts the identity linking mechanism registry. The breaking change that introduced extensible authentication providers is being reconsidered. Implementers should pause identity extensibility work.

OpenClaw Provider Runtime Refactoring: Shared Activation Context

10+ commits from Peter Steinberger and Vincent Koc refactor how provider plugins share context, handle caching, and manage test isolation. Anthropic cache-ttl fix reduces API costs; WhatsApp lazy-loading improves startup time.

UCP Breaking Change: Currency Now Required on Order Schema

Currency becomes a required field on Order, completing a planned migration that ensures capabilities function independently. Orders are now self-describing without Checkout context lookups.

UCP Order Capability Restructuring: Multi-Checkout Support and Signed Quantities

The most significant Order capability evolution since launch — Orders can now reference multiple Checkout sessions, enabling order edits and exchanges with signed quantities for returns.

UCP Breaking Change: Totals Now Use signed_amount

A breaking schema change enables negative totals in UCP checkout flows — unlocking store credits, promotional balances, and refund scenarios that were technically unsupported. Migration required for all implementations.

A2A SDK Contribution Playbook: Community SDKs Get Official Guidance

Two PRs formalize how the A2A ecosystem handles community SDK contributions — verification badges, testing requirements, and maintenance expectations for third-party implementations.

MCP AWT: AI-Powered E2E Testing Server

A new MCP server proposal introduces AI Watch Tester — a framework for AI-powered end-to-end testing of web applications and agent workflows. Visual assertions, self-healing selectors, and behavior verification for the first testing-focused community server.

OpenClaw WhatsApp & Media Stability Sprint

Today's OpenClaw activity focused on WhatsApp channel reliability and media handling. 10+ commits fixing runtime seams, plugin route scopes, session key bindings, and media read helpers for improved stability.

MCP Signet: Cryptographic Action Signing for Tool Calls

A new proposal introduces Signet middleware for cryptographically signing MCP tool call requests and responses. Ed25519 signatures with DID-based identities enable non-repudiable audit trails for enterprise AI deployments.

UCP Identity Linking Mechanism Registry — Breaking Change

A breaking redesign introduces a mechanism registry and capability-driven scopes for identity linking. OAuth, passkeys, and wallets now have explicit trust levels. Migration window through June 2026.

OpenClaw Provider Policy Migration — Plugins Complete Their Takeover

Peter Steinberger completes the multi-week effort to externalize provider-specific policy from the core gateway. Transport hooks generalized, bundled facades cleaned up, and the plugin SDK verified for third-party provider development.

OpenClaw Session Key Canonicalization — Fixing Orphaned Transcripts

A subtle bug caused conversation transcripts to be orphaned when using non-default agent configurations. The fix includes automatic startup migration to recover existing orphaned sessions.

UCP Universal Appointment Scheduling — Services Vertical Arrives

A comprehensive proposal expands UCP beyond physical goods with a three-step workflow for temporal resources: SearchAvailability, HoldSlot, BookAppointment. Distributed leases with TTL prevent race conditions. Includes working MCP server prototype.

Local MCP: Privacy-First Productivity on macOS

A new MCP server brings Mail, Calendar, Contacts, Teams, and OneDrive to AI assistants — all running locally with zero cloud processing. Teams access reads IndexedDB directly, no Microsoft API needed.

UCP Razorpay UPI Payment Handlers — India's Payment Rails Meet Agentic Commerce

Four PRs from Razorpay introduce UPI payment handlers to UCP, including UPI Circle for delegated payments — enabling AI agents to transact within user-set limits. First major Indian payment processor contribution.

OpenClaw Memory Embedding Refactor — Generic Adapters Enable Provider Flexibility

Memory embedding system gets a significant architectural refactor, making adapters provider-agnostic. Hot-swap between OpenAI, Cohere, or local models without touching memory logic.

A2A SDK Reaches v1.0 — Python Tutorial Overhaul

The Agent-to-Agent protocol's Python SDK hits its first stable milestone, with updated tutorials reflecting the modern event-driven approach and native protobuf construction.

OpenClaw Sandbox Policy Fix — alsoAllow and Re-Allow Now Work

A comprehensive fix ensures sandbox tool policy configuration matches runtime behavior. The alsoAllow option was silently ignored, and explicit re-allows of default-denied tools didn't work. Now unified with 815 lines of hardening.

UCP Totals Now Accept Signed Amounts

A subtle schema fix enables negative totals in checkout flows — unlocking discount, credit, and promotional scenarios that were technically unsupported despite being documented in examples.

UCP Identity Linking Redesigned with Mechanism Registry

A significant architectural overhaul transforms identity linking from hardcoded OAuth 2.0 to an extensible mechanism registry pattern. Capability-driven scope derivation makes over-permissioning impossible by construction.

MCP Git Server Closes Argument Injection Gaps

A targeted security fix adds consistent defense-in-depth guards to four git server functions that lacked input validation for leading dashes, closing potential command injection vectors.

UCP CI Pipeline Overhaul and Navigation Fixes

Documentation infrastructure gets a comprehensive cleanup — CI pipeline rebuilt, anchor links fixed throughout the spec, and UI navigation refined. As UCP adoption grows, documentation quality becomes a competitive advantage.

OpenClaw Preserves Anthropic Thinking Block Order

Extended thinking blocks maintain their semantic order through the gateway — critical for debugging agent decisions, building audit trails, and understanding multi-step reasoning processes.

UCP Wallet Attestation: Crypto Identity Meets Traditional Commerce

Wallet-based authentication becomes a first-class identity mechanism in UCP. Third-party attestation providers sign proofs; businesses verify via JWKS offline. Token-gated commerce and DAO membership benefits become practical.

OpenClaw Plugin Runtime Performance: Startup Gets Faster

Multiple commits target plugin runtime performance — lazy loading, parallel initialization, and CI improvements compound for noticeably snappier daily use.

UCP Order Currency Required: Multi-Currency Commerce Gets Explicit

Breaking change makes the currency field mandatory in Order schemas — eliminating ambiguity in cross-border AI agent transactions. Agents will always know what currency they're transacting in.

MCP SQLite Server Gets Critical SQL Injection Fix

Security fix patches SQL injection vulnerabilities in describe_table and query tools. Part of ongoing MCP server hardening as implementations move into production use.

UCP Identity Linking Mechanism Registry Merges: Extensible Authentication Arrives

Breaking change replaces hardcoded OAuth providers with a discoverable registry pattern. Hours later, wallet_attestation lands as the first new mechanism type — enterprise SSO and crypto wallets now have a path.

OpenClaw Matrix Encryption Isolation: Federated E2EE Gets Production-Ready

Coordinated commits harden Matrix integration — runtime encryption loading, isolated credential writes, and session binding fixes. Federated AI assistants with end-to-end encryption mature.

OpenClaw Plugin SDK npm Publish Workflow: Third-Party Plugins Go Public

Official tooling for plugin authors to version, build, and publish to npm. The ecosystem matures from internal-use to open contribution with automated CI, contract validation, and registry sync.

UCP Structured Inapplicable Discount Codes: Teaching AI Agents Why Coupons Fail

New schema for machine-readable discount failure reasons — enabling agents to understand WHY codes don't apply and take corrective action. Minimum not met? Agent suggests items to add.

UCP Totals Contract Formalization: How Order Sums Really Work

Explicit definitions for subtotal, total, and grand_total — formalizing calculation relationships and validation rules. Essential infrastructure for enterprise adoption where audit trails and compliance matter.

OpenClaw Exec Sandbox Security: Blocking Environment Injection Vectors

Critical security fix blocks JVM, Python, and .NET environment variable injection attacks that could bypass sandbox restrictions. JAVA_TOOL_OPTIONS, PYTHONSTARTUP, and DOTNET_STARTUP_HOOKS now filtered.

UCP Identity Linking Gets a Complete Redesign

Breaking change proposes mechanism registry pattern and capability-driven scopes — replacing hardcoded identity providers with a discoverable, extensible system. Essential for enterprise compliance and future DID support.

OpenClaw Deferred Channel Startup: Faster Boots, Same Reliability

Coordinated commits restructure gateway startup to accept HTTP requests immediately while channels connect in the background. Boot times drop from 30s+ to <500ms in flaky network conditions.

UCP Trust Infrastructure Matures: Eligibility, Attestation, and Totals

Three PRs this week advance enterprise trust: eligibility verification merged, cryptographic attestation proposed, and totals contracts formalized. Together they answer how businesses can trust AI agents.

OpenClaw March Stability Sprint: WhatsApp, Android, and Gateway Hardening

Over a dozen fixes in 48 hours address production edge cases: WhatsApp timestamp precision, Telegram HTML chunking, Android call log search, and remote CDP probes.

UCP Authorization & Abuse Signals: Platform Attestations for Commerce Security

Breaking change enables fraud prevention — Ilya Grigorik's PR introduces platform-attested signals for authorization, rate limiting, and abuse prevention. Businesses can now request environmental data that platforms observe directly.

OpenClaw Wear OS MVP: Your AI Assistant on Your Wrist

sibbl delivers a complete Wear OS companion app — transcript-first chat UI, direct/proxy gateway connectivity, tiles, complications, and background reply handling. Conversational AI designed for tiny round screens.

A2A 1.0 Official Release: Agent Interoperability Goes Production

After six months of iteration, Google's Agent-to-Agent Protocol reaches 1.0 with specification freeze, blog announcement, and semantic versioning commitments. Enterprise adopters now have API stability guarantees.

MCP Security Fixes: Git Argument Injection and Robots.txt Edge Cases

Elliot lands two security fixes: argument injection guards for git server and fail-closed handling when robots.txt returns 5xx server errors.

UCP Return Extension: Completing the Commerce Loop

A new extension enables AI agents to understand return policies, methods, and fees — answering "Can I return this?" before purchase and enabling true cost transparency.

OpenClaw Security Sprint: Unicode Obfuscation, WebSocket Limits, Webhook Hardening

Vincent Koc lands six security-focused commits addressing command injection vectors, pre-auth resource exhaustion, and webhook authentication gaps.

A2A Extension Governance: Formalizing the Extension Lifecycle

Vinay Ramesh establishes formal governance for A2A extensions — lifecycle stages, TSC oversight, and compatibility requirements as the protocol enters enterprise-ready phase.

UCP Eligibility Claims: Verified Discounts for AI Agents

Ilya Grigorik proposes cryptographic eligibility claims — enabling AI agents to prove customer eligibility for restricted discounts (student, employee, membership) without exposing identity.

UCP: Extending Discount Capability to Cart

Ilya Grigorik extends UCP's discount capability to cart operations — discount codes applied early MUST persist through checkout, enabling natural shopping flows for AI agents.

OpenClaw: Stripping Leaked Model Control Tokens

George Zhang adds generic pattern matching to sanitize internal delimiter tokens from GLM-5, DeepSeek, and other providers — cleaner multi-provider output for end users.

OpenClaw 2026.3.9 Release Prep: LaunchD Hardening and Build Stability

Peter Steinberger lands 10 commits preparing the release — hardening macOS launchd permissions, stabilizing paths, and fixing onboarding edge cases.

A2A RELEASES.md: Formalizing the Release Process

Following the AWS TSC addition, Amye Scavarda Perrin opens PR #1605 documenting versioning, deprecation, and upgrade guidance — essential infrastructure for enterprise adopters.

UCP Catalog Capabilities: Product Discovery for AI Agents

PR #55 adds Catalog Search and Catalog Lookup operations — agents can now discover products directly through the protocol, completing the end-to-end commerce loop.

OpenClaw Talk Mode Improvements: Configurable Silence Detection

A flurry of commits hardens voice conversation handling across iOS and macOS — configurable silence timeouts, shared config parsing, and better edge case handling.

UCP Version Negotiation Overhaul: Business-Canonical Versioning

Ilya Grigorik's comprehensive PR #200 fundamentally reshapes how protocol versioning works — businesses declare what they support, platforms must match, and extensions can declare dependency constraints.

A2A Governance Matures: AWS Joins Technical Steering Committee

The addition of a new AWS representative to A2A's TSC signals enterprise adoption momentum — Google's agent interoperability protocol is becoming an industry standard, not a single-vendor play.

A2A Pub/Sub Proposal: Event-Driven Agent Coordination

As multi-agent systems scale beyond simple request-response patterns, native publish/subscribe semantics become essential infrastructure for event-driven agent coordination.

OpenClaw HEIC Normalization: Cross-Platform Image Handling

A small gateway fix solves a frustrating problem: iOS photos can now be processed by any backend, not just Apple-native ones. Gateway-level normalization benefits all downstream consumers.

A2A AIAR Identity Extension: First Agent Registry MVP

After weeks of theoretical proposals, someone ships working code — the AIAR extension provides a concrete foundation for agent verification with registry-based identity, capability attestations, and trust scoring.

OpenClaw Mattermost Interactive Buttons: Enterprise Messaging Expands

Mattermost becomes the third major messaging platform with rich UI support in OpenClaw — interactive buttons enable approval workflows, quick actions, and guided interactions for self-hosted enterprise deployments.

A2A Push Notification Consolidation: Simplifying Agent Communication

Google merges TaskPushNotificationConfig and PushNotificationConfig into a single type — reducing cognitive overhead for implementers building real-time agent systems.

OpenClaw Sessions Spawn Streaming: Fixing Sub-Agent Communication

A critical fix ensures parent sessions receive streaming output from spawned sub-agents via ACP relay — enabling real-time visibility into background agent work.

A2A Agent Passport Implementation: From Proposal to Production

A community member ships a working agent identity system — Ed25519 cryptographic identity, scoped delegation with cascade revocation, and automatic enforcement. Addresses #1497, #1472, and #1501 simultaneously.

UCP Order Currency Field: Multi-Currency Agentic Commerce Gets Cleaner

A top-level currency field arrives in the Order schema — simplifying multi-currency detection, comparison, and display for AI agents shopping across borders.

UCP Ancillaries Extension Proposal: Add-Ons for Agentic Commerce

Structured support for optional add-ons, upgrades, and cross-sells becomes first-class in the Universal Commerce Protocol — essential infrastructure for travel, hospitality, and e-commerce.

OpenClaw Security Hardening: Webhook Routes, SMS, and Plugin Isolation

Peter Steinberger lands six security-focused commits addressing webhook authentication, HTTP handler isolation, and SMS sending defaults — continuing systematic hardening.

OpenClaw Slack Stability Overhaul: Production-Ready Enterprise Messaging

Eight Slack-related commits from six contributors — streaming modes, reply routing, agent identity, and startup reliability. The Slack integration transforms from early-adopter to production-ready.

UCP Schema Validation CI: Automated Quality Gates for Agentic Commerce

Three PRs add comprehensive CI infrastructure — schema validation workflows, pre-commit sync checks, and improved path triggers. Quality gates for schema-driven development.

OpenClaw Android Voice Reliability: Production-Grade Voice Mode Arrives

Six commits land addressing token rotation, retry logic, mute handling, and conversation state — transforming voice from demo feature to daily driver.

UCP Status Discriminator: Semantic Error Handling for Agentic Commerce

Ilya Grigorik proposes structured status fields with unrecoverable severity — enabling AI agents to make intelligent decisions about checkout failures.

OpenClaw Auto-Enables OpenAI Responses Compaction: Context Window Optimization

Server-side compaction arrives by default — intelligently summarizing older conversation turns to preserve context window space. Cost reduction and better UX for always-on assistants.

A2A Bidirectional Streaming Proposal: gRPC Gets Native Agent Communication

A breaking change proposal for native streaming over gRPC — sub-millisecond latency versus polling. Could reshape how autonomous agents collaborate in real-time.

OpenClaw Gateway Security: Path Canonicalization and Exec Approvals

Peter Steinberger lands six commits on gateway security — shared path canonicalization, plugin route auth hardening, and exec approvals refactoring. Defense in depth for AI assistants handling sensitive operations.

MCP Deterministic Capability Ordering: Why Tool Registration Order Matters

A pair of contributions address protocol predictability: when servers expose capabilities during initialization, the order should be deterministic. Essential for testing, caching, and debugging.

OpenClaw Android Voice Mode Redesign: Full-Height Conversations Arrive

Ten commits transform voice UX — full-height conversation layout, live transcript streaming, and improved state management. Voice-first interaction matures from novelty to primary input method.

A2A DID-Based Agent Identity: Decentralized Verification via AIP

A new AIP proposes using W3C Decentralized Identifiers for agent identity — giving autonomous agents verifiable, self-sovereign credentials with cryptographic proof and delegation chains.

UCP Payment Instrument Qualifiers: Enabling BNPL and Installments

A new field enables AI agents to communicate payment method capabilities like "can do 4-installment BNPL" — unlocking promotional financing and installment plans in agentic commerce.

MCP Declaration Validation: Reference Servers Adopt "Fail Closed" Posture

Unknown declaration entries now fail startup deterministically across core MCP servers — a security hardening that catches misconfigurations before production.

OpenClaw Routes Claude Through Google Vertex AI: Enterprise Multi-Cloud Arrives

Claude models can now route through Google Cloud's infrastructure — keeping traffic within GCP, consolidating billing, and enabling enterprise deployment patterns.

A2A Renames "Blocking" to "Polling": Why Terminology Matters in Protocol Design

Post-V1.0 cleanup continues: the confusingly-named "blocking" parameter becomes "polling" — because it describes client-side polling, not server-side blocking.

OpenClaw Core Auto-Updater: Keeping AI Assistants Current

Peter Steinberger adds a core auto-updater with dry-run preview — addressing the operational challenge of keeping always-on AI assistants running the latest code without breaking active sessions.

OpenClaw Multi-Language Memory Search: Japanese, Spanish, Portuguese Support

Vincent Koc adds Japanese, Spanish, and Portuguese query expansion to OpenClaw's full-text search — enabling memory recall to work properly across languages by filtering language-specific stop words.

OpenClaw Discord Thread-Bound Subagents: AI Teams Get Their Own Workspaces

A significant UX improvement: sub-agents spawned for specialized tasks now get their own Discord threads. Messages route to dedicated workspaces, bindings persist across restarts, and cleanup is automatic.

UCP Payment Instruments: First-Class Support for Payment Method Constraints

A proposal adds available_instruments to payment handler configurations — enabling handlers to declare supported payment methods with constraints. Essential infrastructure for installments and BNPL.

UCP Catalog Lookup: Direct Product Retrieval for AI Agents

A new proposal adds a get product operation to catalog.lookup — enabling AI agents to retrieve specific products by ID rather than searching. Small API addition, big efficiency gains.

OpenClaw iOS TLS Hardening: Forcing Encryption for Remote Gateways

A security fix forces TLS for all non-loopback gateway connections — closing a gap where manual configuration could allow unencrypted remote traffic.

MCP Filesystem Server Gets Cross-Platform Path Fixes

Two important fixes for production deployments: macOS symlink resolution and Windows absolute path parsing. As MCP servers move into enterprise use, these platform-specific edge cases matter.

UCP Message Signing Merges: Cryptographic Trust Arrives

After weeks of refinement, Ilya Grigorik's comprehensive signing specification lands — RFC 9421 HTTP Message Signatures, JWK key discovery, and profile-based trust for enterprise agentic commerce.

A2A Agent Identity and Trust Framework Proposal

A comprehensive proposal addresses the "who are you?" problem for autonomous agents — proposing verification levels, trust signals, and delegation chain security.

UCP Ancillaries Extension: Add-Ons Get First-Class Protocol Support

The second vertical extension in as many days shows UCP's extensibility model is gaining momentum — now tackling optional add-ons, upgrades, and upsells.

A2A Adds API Linter to CI: Protocol Quality Gets Automated

As the Agent-to-Agent Protocol matures past 1.0, Google adds automated proto file validation — a sign the spec is ready for enterprise implementation.

UCP-Hotels Extension: Sabre Brings Travel GDS Expertise to Agentic Commerce

The protocol's first vertical extension arrives from a major travel technology company, signaling UCP's extensibility model is working as intended.

OpenAttribution Telemetry: Tracking Content Influence in AI Commerce

A comprehensive spec for attributing purchases to the content that influenced them — solving the "who gets credit" problem in AI-assisted shopping.

OpenClaw's Post-Security Refactoring Wave

One day after the Valentine's Day security blitz, Peter Steinberger pivots to a systematic deduplication effort — consolidating session handling, authentication, and platform integrations.

UCP Introduces AI-Powered Repository Workflows

The Universal Commerce Protocol repository adds four AI-powered GitHub workflow agents for PR triage, issue management, discussion moderation, and metrics reporting.

OpenClaw Valentine's Day Security Blitz: 20+ Fixes in a Single Day

Peter Steinberger leads a coordinated security hardening effort, addressing webhook routing, shell injection, TLS pinning, and process isolation.

UCP Attribution Signals Extension: Tracking Conversions in Agentic Commerce

A new Enhancement Proposal addresses a critical gap: when an AI agent completes a purchase, how does the merchant know which touchpoints influenced the conversion?

A2A Cryptographic Identity Verification: Trust at the Protocol Layer

A new proposal addresses a fundamental gap: how do autonomous agents verify each other's identity before delegating sensitive tasks?

OpenClaw Canvas Security: Restricting IP-Based Auth to Private Networks

A security fix closes a potential attack vector in OpenClaw's canvas feature, part of a coordinated hardening effort.

A2A Education Goes Mainstream: DeepLearning.AI Course

Google and DeepLearning.AI partner on a comprehensive A2A course, signaling the protocol has matured enough for structured education.

MCP Memory Server: Exposing Knowledge Graphs as Resources

A new proposal shifts MCP memory from tool-based to resource-based access — enabling fundamentally different interaction patterns.

MCP Security Audits Reveal Vulnerabilities Across Core Servers

SQL injection in SQLite, path traversal in filesystem, and memory handling issues — as MCP adoption grows, security researchers are finding real problems.

A2A Extension Governance Framework: How Should Agent Protocols Evolve?

A new proposal asks the fundamental question: as A2A matures past 1.0, how should the community manage extensions without fragmenting the ecosystem?

OpenClaw Adds Custom Provider Onboarding: Local AI Gets First-Class Support

A new configuration flow makes it dramatically easier to connect self-hosted and custom AI providers. For privacy-conscious users and enterprises, this removes a significant barrier.

A2A 1.0: What Changed from v0.3.0 and Why It Matters

Google's Agent-to-Agent Protocol reaches its 1.0 milestone with a comprehensive migration guide for implementers.

UCP Endorsed Partners: Block, Fiserv, Klarna, Splitit Join the Ecosystem

Four major payment infrastructure companies commit to the Universal Commerce Protocol, signaling enterprise confidence in agent-driven transactions.

OpenClaw Security & Trust Documentation: The Enterprise Readiness Signal

As AI assistants move from side projects to enterprise infrastructure, dedicated security documentation signals a project's maturity.

MCP Everything Server Gets Comprehensive Tests: Ecosystem Maturity in Action

When a reference implementation gets proper test coverage, it signals the ecosystem is ready for production.

A2A Agent-Mesh Trust Layer: Cryptographic Identity for Autonomous Agents

A new contribution proposes decentralized identity verification for A2A agents — addressing how autonomous systems can verify each other's identity before delegating sensitive tasks.

OpenClaw Adds Baidu Qianfan: Expanding Global AI Provider Support

The first major Chinese AI platform joins OpenClaw's provider roster, marking continued expansion of the project's global reach.

OpenClaw Security Hardening: From Hobbyist to Enterprise-Ready

Five security PRs in one day signal a coordinated push to address vulnerabilities before they become incidents. As AI assistants handle increasingly sensitive data, this matters.

UCP Request Signing: Cryptographic Trust for Agentic Commerce

As AI agents prepare to handle real money, Google proposes cryptographic signing for UCP requests and responses — laying the foundation for verifiable, tamper-proof transactions.

A2A 1.0 Release Candidate: The Final Push

Google's Agent-to-Agent Protocol approaches its 1.0 milestone with a flurry of spec refinements — clarifying timestamps, simplifying IDs, and enabling SDK backwards compatibility.

UCP Risk & Abuse Signals: Building Trust Infrastructure for Agentic Commerce

A new proposal brings fraud prevention and rate limiting primitives to the Universal Commerce Protocol — essential infrastructure for enterprise adoption.

UCP Intent Field: Teaching AI Agents to Communicate Shopping Context

A new field enables AI agents to express buyer intent in natural language — unlocking personalization without authentication.